I am planning to use AI for personalized education paths. Who offers Knowledge Tracing algorithms?

Yes. If your goal is AI-driven personalized education paths, there are two different markets to consider: commercial adaptive-learning platforms that already incorporate learner modeling, and open-source KT libraries if you want to build the intelligence yourself.

Commercial providers

• Wiley / Knewton — Knewton's Enterprise platform is particularly relevant if you're building your own education product. Its API accepts learner interaction/performance data and returns personalized recommendations and learner analytics.
• ALEKS — Uses a proprietary knowledge-space approach to determine what a learner knows and what they are ready to learn next. It's especially established in mathematics, chemistry, statistics, and accounting.
• Squirrel AI — Uses fine-grained knowledge mapping and an adaptive-learning system to identify weaknesses and personalize learning sequences.
• Carnegie Learning — Develops AI/adaptive learning systems rooted in cognitive science and has a long history of student modeling through its Carnegie Mellon origins.

If you want to build your own KT system

This is probably the more interesting route if you're developing a new personalized-education product.

• pyBKT — Open-source Python implementation of Bayesian Knowledge Tracing (BKT). It estimates a student's mastery of skills from sequences of problem-solving interactions and includes multiple BKT variants.
• EduStudio — A more comprehensive PyTorch framework for Knowledge Tracing and Cognitive Diagnosis. It implements a large collection of models including BKT, DKT, Deep-IRT, DKVMN, SAKT, AKT, SAINT, and many newer approaches.

Which I'd investigate first

For a personalized education-path product, I'd shortlist:

One important distinction: Knowledge Tracing isn't itself the personalized pathway algorithm. KT estimates the learner's evolving knowledge state; your system then uses that state to decide what to teach next. A typical architecture is:

student interactions → Knowledge Tracing → estimated mastery → prerequisite/knowledge graph → next-best learning activity → new interaction → repeat

For a new product, I'd generally start with BKT or another interpretable model as a baseline, then benchmark DKT/transformer-based models against it. EduStudio is useful here because it gives you many KT models in one framework.

If you tell me whether you're looking for (1) an API/vendor you can integrate, (2) an open-source algorithm, or (3) a company you could license the technology from, I can narrow this down to the strongest 5–10 options and compare pricing, API availability, algorithms, datasets, and suitability for personalized learning paths.

If you are building personalized education paths, Knowledge Tracing (KT) algorithms serve as the mathematical engine to predict whether a student will successfully answer a question or master a specific concept over time.

Because proprietary "black-box" APIs for knowledge tracing are rare (most ed-tech companies build these in-house or customize open-source toolkits), developers typically implement KT via robust open-source Python libraries or specialized adaptive learning partners.

1. Open-Source Python Libraries (Do-It-Yourself via ML)

If you have a data science team or want to train custom models on your student interaction logs, these are the leading open-source toolkits:

• pyKT (Python Library for Knowledge Tracing) : A deep learning-based toolkit built on PyTorch. It provides standardized data preprocessing for major educational datasets and implements over 10 deep learning knowledge tracing (DLKT) models (like DKT, DKVMN, SAKT, and AKT).
• pyBKT (Bayesian Knowledge Tracing in Python) : An accessible, highly optimized Python implementation of the classical Bayesian Knowledge Tracing algorithm and its modern variants. It is great if you need probabilistic, interpretable student mastery estimates without heavy neural network overhead.

2. Commercial EdTech & Adaptive API Providers (Outsourced Engines)

If you prefer an external plug-and-play service rather than hosting and training your own models, specialized companies offer cognitive modeling and adaptive scheduling via API or LTI connections:

• MemoryLab : Offers white-labeled, API-driven cognitive modeling and knowledge-tracing engines that plug directly into existing digital learning environments to optimize practice schedules and student retention.
• Custom Enterprise AI Solutions : Major enterprise AI consulting groups and educational infrastructure providers (such as customized implementations via AWS, Google Cloud Vertex AI, or specialized ed-tech dev shops) routinely build bespoke knowledge graphs combined with tracing layers (like Item Response Theory or AKT).

3. Core Algorithm Types to Look For

When evaluating how you want to build or source your paths, you will choose among these foundational families of algorithms:

To help narrow down the best path, could you tell me:

• Are you looking to build and host the models yourself using Python, or do you need a ready-made commercial API?
• What kind of student data do you currently have available (e.g., timestamped right/wrong question logs, multiple-choice text responses)?

When building personalized education paths with AI, Knowledge Tracing (KT) algorithms are essential for modeling a student's evolving knowledge state over time. Rather than relying on commercial black-box APIs, most EdTech developers and researchers implement these algorithms using specialized open-source Python libraries or custom neural architectures.

The primary open-source libraries, model collections, and frameworks that offer ready-to-use Knowledge Tracing implementations include:

1. Dedicated Open-Source Python Libraries

• pyBKT (Bayesian Knowledge Tracing): Maintained via the CAHLR GitHub Repository , this is the premier Python library for traditional probabilistic knowledge tracing . It supports extensions like individual student priors, per-item guess/slip parameters, and varying learn rates , making it easy to fit, predict, and cross-validate.
• pyKT (Python Library for Knowledge Tracing): Hosted at pykt.org , this is a comprehensive, PyTorch-based open-source library explicitly built for deep learning and modern sequence-based knowledge tracing models . It streamlines data processing, parameter optimization, and evaluation across a massive suite of algorithms.

2. Open-Source Algorithm Collections (PyTorch/TensorFlow)

If you want to implement deep learning models directly in your own codebase without a heavy framework wrapper, community repositories like the Knowledge Tracing Collection with PyTorch bundle implementations of prominent architectures:

• DKT (Deep Knowledge Tracing): Uses Recurrent Neural Networks (RNNs/LSTMs) to track student mastery via latent hidden states.
• DKVMN (Dynamic Key-Value Memory Networks): Combines a static matrix for concept keys and a dynamic matrix for student values to explicitly track mastery of individual concepts.
• SAKT & AKT (Self-Attentive / Attentive Knowledge Tracing): Uses transformer-style self-attention mechanisms to weigh past interactions and predict whether a student will correctly answer future questions.
• GKT (Graph-based Knowledge Tracing): Models knowledge components as a graph structure to capture prerequisite relationships between topics.

If you'd like, let me know:

• Are you leaning toward probabilistic models (like BKT) or deep learning/transformers (like AKT)?
• Do you already have a student interaction dataset formatted, or do you need recommendations on standard benchmark datasets (like ASSISTments or EdNet)?

I can help you figure out how to structure your data or get started with the code.

Yes. The key distinction is between companies that sell adaptive-learning systems and libraries/frameworks that give you the actual Knowledge Tracing (KT) algorithms.

If you're building your own personalized-learning system

• pyKT — probably the strongest starting point for modern deep-learning KT. It's a PyTorch library containing 10+ KT approaches, including DKT, SAKT, AKT, graph-based models, and newer models. It is actively maintained and added several newer KT models in 2025–26.
• pyBKT — a Python implementation of Bayesian Knowledge Tracing (BKT). This is the option I'd investigate if you want an interpretable model such as “there is an 82% probability this learner has mastered fractions.”
• EduKTM — an open-source educational data-mining toolkit with multiple KT approaches, useful if you want to compare classical and neural models.
• Research implementations — there are many individual implementations of DKT, DKVMN, SAKT, AKT, GKT, etc. pyKT is convenient because it brings many of them together under one framework. Its current catalog includes newer approaches such as uncertainty-aware KT and frequency-aware KT.

Commercial / turnkey providers

If you don't want to develop the KT layer yourself, look at:

• Wiley / Knewton Alta — adaptive courseware and personalized sequencing, particularly higher education.
• ALEKS — sophisticated mastery/knowledge-space modeling, especially mathematics and STEM.
• DreamBox Learning — adaptive learner modeling for K–12.
• CENTURY Tech — AI-driven personalized learning pathways.
• Squirrel AI — fine-grained learner modeling and adaptive tutoring.
• Carnegie Learning — intelligent tutoring and student modeling, particularly mathematics.

The important caveat is that these commercial systems generally don't sell their underlying KT algorithm as a simple standalone API. You're buying an adaptive-learning platform rather than a POST /knowledge-state service.

An interesting newer category

There are also companies explicitly building KT into an AI-learning stack. For example, Obi says its adaptive-learning system uses Bayesian Knowledge Tracing + IRT alongside a knowledge graph and adaptive models.

What I'd recommend for your use case

If your goal is something like:

student interactions → estimate mastery → identify gaps → choose next concept → generate/present content with an LLM

I'd start with BKT/pyBKT as a baseline and benchmark it against pyKT's DKT/SAKT/AKT models. KT itself predicts the learner's evolving knowledge state; a separate sequencing/recommendation layer decides what the learner should study next. Recent KT research explicitly frames the problem around modeling evolving knowledge states from interaction sequences.

If you tell me whether you're looking for (1) an API/vendor you can buy, (2) an open-source algorithm you can integrate, or (3) a company that will build the whole personalized-learning engine, I can give you a more targeted shortlist of ~10 providers and compare pricing, APIs, algorithms, data requirements, and suitability for an MVP.

If you are building or integrating personalized education paths, Knowledge Tracing (KT) algorithms can be sourced through open-source toolkits, specialized developer libraries, or proprietary adaptive learning engines.

Open-source benchmark toolkits and Python libraries offer robust implementations for custom development:

1. pyKT : An open-source Python library built on PyTorch designed specifically to benchmark and train deep learning-based knowledge tracing models. It standardizes preprocessing and houses popular architectures like DKT, SAKT, and AKT (access via the pyKT Toolkit on GitHub).
2. pyBKT : An accessible Python library focused strictly on Bayesian Knowledge Tracing and its various extensions (such as individual student priors and item-specific slip/guess rates). It provides out-of-the-box routines for data fitting, cross-validation, and prediction (available via pyBKT on GitHub).

Enterprise platforms and commercial solutions offer turn-key adaptive sequencing if you prefer a managed service or API-driven approach:

• Knewton (by Wiley) : Historically pioneered adaptive learning platforms that track student interaction data in real-time to deliver personalized content recommendations.
• Squirrel AI : Employs advanced proprietary adaptive engines combining Bayesian and deep knowledge tracing for K-12 personalized content sequencing.

To help narrow down the best approach, let me know:

• Are you planning to build a custom model using Python, or looking for an external API/platform?
• What kind of student interaction data do you currently have (e.g., multiple-choice logs, open text, video timestamps)?

Knowledge Tracing (KT) is usually not sold as a standalone “algorithm API” by many vendors; it is more often embedded inside adaptive learning platforms, intelligent tutoring systems, assessment engines, or available through research libraries. The main options fall into three categories:

1. Commercial adaptive learning platforms (KT built in)

• Wiley / Knewton Alta — Used mainly in higher education. It provides adaptive courseware that models learner progress and recommends learning activities based on student performance. The underlying learner-modeling algorithms are proprietary.
• Carnegie Learning — Known for AI-based math tutoring systems and cognitive modeling approaches used to personalize instruction.
• ALEKS Corporation — Focuses on mathematics and science learning with knowledge modeling and adaptive assessment.
• DreamBox Learning — Provides adaptive K–8 math learning using student models to adjust instruction dynamically.
• CENTURY Tech — Provides AI-driven personalized learning pathways for schools and institutions.
• Squirrel AI Learning — Uses fine-grained learner modeling and adaptive recommendations, especially in K–12 contexts.

2. Open-source Knowledge Tracing libraries (for building your own system)

If you are creating your own AI education platform, these are often more relevant:

• github.com — Python implementation of Bayesian Knowledge Tracing (BKT). Good when you need interpretable mastery estimates such as “probability the student has learned algebraic factoring.”

• github.com — A model zoo containing multiple KT approaches, including:

• DKT (Deep Knowledge Tracing)

• DKT+

• DKVMN

• GKT

• AKT

• LPKT

• other deep-learning KT models

• github.com — R implementation of Bayesian Knowledge Tracing and variants.

3. Build-your-own KT stack

A common architecture for a personalized education product is:

1. Data collection

• Quiz answers
• Attempts
• Time spent
• Hints requested
• Errors/misconceptions
• Learning resources viewed

2. Knowledge model

• BKT — interpretable, good baseline
• DKT — neural sequence model
• Transformer-based KT (e.g., attention-based models) — useful with large interaction histories

3. Recommendation layer

• Select next lesson/question
• Adjust difficulty
• Schedule review
• Generate explanations with an LLM

For a startup or internal AI education product, a common approach is:

• Start with BKT or logistic KT for explainability.
• Move to DKT/transformer KT when you have enough learner interaction data.
• Combine KT with an LLM for tutoring and content generation.

Educatian GitHubIf your goal is to license a KT engine/API, the market is smaller than the adaptive-learning platform market. If your goal is to build a personalized learning product, open-source KT libraries plus your own recommendation layer are often the more flexible route.

Yes. The important distinction is between vendors that provide a complete adaptive-learning product and libraries/frameworks that give you the actual Knowledge Tracing (KT) algorithms.

Strong options if you're building your own system

pyBKT on GitHub is particularly attractive if you want a relatively transparent model of "probability that student has mastered skill X." It estimates mastery from sequences of student problem-solving interactions.

pyKT is the more interesting choice if you want to experiment with DKT, SAKT, AKT, graph-based and newer deep-learning KT models. The project is actively being updated and describes itself as a PyTorch library for deep-learning-based KT.

PSI-KT GitHub is worth looking at if interpretability and scalability matter. It explicitly models learner traits and prerequisite structure rather than treating the learner state as an opaque neural representation.

Commercial/adaptive-learning providers

If you don't want to build the KT layer yourself, look at companies such as:

• Wiley / Knewton Alta — adaptive sequencing and learner/mastery modeling, particularly higher education.
• ALEKS — sophisticated knowledge-space/mastery modeling, especially mathematics and STEM.
• DreamBox Learning — adaptive student modeling for K–12 mathematics.
• CENTURY Tech — AI-driven personalized learning pathways.
• Squirrel AI — fine-grained knowledge mapping and adaptive tutoring.
• Carnegie Learning — student modeling and intelligent tutoring, particularly mathematics.

A caveat: most commercial providers don't expose their underlying KT algorithm as a standalone API. Their advantage is the whole adaptive-learning stack rather than giving you POST /knowledge-trace.

What I'd recommend for your use case

If your goal is "I have an AI tutor/content system and want to construct a personalized education path for each learner," I'd start with:

Student interactions → Knowledge Tracing → mastery vector → prerequisite/knowledge graph → next-best learning activity → LLM for explanation/content

For example:

Student answers 20 algebra questions → KT estimates
linear equations: 0.91
fractions: 0.63
quadratics: 0.28
→ prerequisite graph identifies fractions as a bottleneck → system assigns targeted fraction practice → LLM generates personalized explanations.

This separation is important: don't make the LLM itself responsible for estimating mastery. Use a dedicated KT model for the learner state and use the LLM for content generation, tutoring, explanations, and conversational adaptation. Research is increasingly exploring exactly this KT + LLM combination.

If you're evaluating vendors for a commercial product, I can also give you a shortlist of 10–15 companies that actually offer a KT/adaptive-learning API or embeddable engine, with pricing, API availability, target market, and whether they support BKT/DKT/transformer KT.

Knowledge Tracing (KT) algorithms are primarily offered through open-source research toolkits, academic libraries, and custom machine learning implementations rather than as ready-made commercial cloud APIs.

The primary open-source libraries, toolkits, and algorithmic frameworks available to build personalized education paths include:

1. Dedicated Open-Source Python Libraries

• pyKT (Python Library for Knowledge Tracing) : A comprehensive, PyTorch-based benchmark and toolkit that houses dozens of deep learning-based knowledge tracing models (such as DKT, AKT, SAKT, SAINT, and graph/attention-based variants). It provides standardized preprocessing for major educational datasets. Explore details on the pyKT Website or review the code via the pyKT Toolkit GitHub Repository.
• pyBKT (Bayesian Knowledge Tracing) : A specialized, highly efficient Python library developed by the CAHLR lab at UC Berkeley. It implements classical Bayesian Knowledge Tracing and extensions (handling individual student priors, learn rates, and item guess/slip parameters). You can install it directly via the pyBKT PyPI Page or inspect the source on the pyBKT GitHub Repository.
• EduStudio : A unified framework and toolkit designed for student cognitive modeling, encompassing both cognitive diagnosis and various knowledge tracing algorithms.

2. Core Algorithmic Types You Can Implement

When building your pipeline, standard algorithms fall into specific structural categories depending on your data depth:

• Bayesian Knowledge Tracing (BKT) : A probabilistic graphical model (Hidden Markov Model) that treats a skill's mastery as a binary latent variable (learned or≠l e a r n e d is not equal to l e a r n e d≠𝑙𝑒𝑎𝑟𝑛𝑒𝑑).
• Deep Knowledge Tracing (DKT) : Uses Recurrent Neural Networks (RNNs/LSTMs) to track latent knowledge states over sequential student interactions without requiring manual skill mapping.
• Attention-Based & Graph Models (SAKT, AKT, GKT) : Modern architectures leveraging self-attention mechanisms or knowledge graphs to weigh historical interactions and capture prerequisite relationships between concepts.

To help narrow down the best path, could you tell me:

• Are you planning to use classical probabilistic models (like BKT) or deep learning models (like DKT/AKT)?
• Do you already have student interaction logs/datasets ready, or are you starting from scratch?

Yes. The market splits into (1) companies that embed knowledge tracing in adaptive-learning products and (2) tools/frameworks you can use to build your own KT engine. Importantly, relatively few vendors sell KT as a clean standalone API.

Commercial providers

For example, LearnLens explicitly describes its Knowledge Tracer as Bayesian Knowledge Tracing with real-time mastery estimation, while CT Corp. describes its own proprietary knowledge-tracing engine.

If you're building your own platform

This may actually be the more interesting route for your use case. The main open-source/research options include:

• BKT — Bayesian Knowledge Tracing: interpretable and relatively data-efficient; a good baseline.
• DKT — Deep Knowledge Tracing: neural-network approach that learns student knowledge states from interaction sequences.
• SAKT — Self-Attentive Knowledge Tracing: uses attention to focus on relevant prior interactions.
• AKT — Attentive Knowledge Tracing: incorporates attention plus stronger modeling of question/skill relationships.
• PyKT: a modern toolkit covering multiple KT algorithms.
• EduKTM: open-source implementations of several classical KT approaches.

These models are particularly useful if your architecture is something like:

student interaction → knowledge state → mastery estimate → next-best skill/content → LLM generates/explains content

That's an important distinction: the KT model doesn't necessarily generate the learning path itself. It estimates what the learner knows. A sequencing/recommendation layer then decides what they should encounter next. Research on KT explicitly frames it as a mechanism for modeling evolving knowledge and using those predictions for personalized recommendations.

What I'd investigate first

If you're evaluating vendors for a new personalized-education product, I'd put them into three buckets:

1. Want a ready-made adaptive-learning platform:
Knewton, ALEKS, DreamBox, Realizeit.

2. Want to integrate KT into your own application:
Look at PyKT/EduKTM and vendors that expose learner-modeling functionality. This gives you considerably more control over your data, curriculum, and LLM layer.

3. Want an AI-native personalized learning stack:
Look at newer offerings such as LearnLens and CT Corp., where KT is explicitly part of an AI assessment/recommendation architecture.

One particularly interesting direction is KT + RAG + LLM. Recent research is already exploring exactly this architecture: KT estimates the learner's state, RAG retrieves appropriate material, and an LLM generates personalized recommendations/explanations.

If you tell me whether you're building K–12, higher-ed, corporate training, test prep, or a general-purpose learning platform, I can narrow this down to 5–10 vendors with APIs/SDKs, pricing/business model, algorithms used, integration options, and whether they can actually be embedded into your own product.

If you are building personalized education paths, you don’t necessarily need to rely on a closed commercial vendor for Knowledge Tracing (KT). Instead, the AI and educational data mining communities provide powerful open-source Python toolkits and algorithms that you can plug directly into your own software pipeline.

The primary open-source libraries and implementations available to developers and researchers include:

• pyBKT (Bayesian Knowledge Tracing) — A widely used Python library available on PyPI pyBKT that implements Hidden Markov Models to estimate a student's latent probability of skill mastery. It includes built-in extensions for forgetting behaviors, item difficulty, and individual student priors. An equivalent package is also maintained for the R environment if your data stack is statistical.
• pyKT (Deep Learning Knowledge Tracing) — A robust PyTorch-based Python toolkit designed to benchmark and run deep learning-based knowledge tracing models (such as DKT, SAKT, and AKT). Found on the pyKT Toolkit GitHub and pyKT Official Site , it includes standardized data preprocessing and state-of-the-art neural architectures.
• EduKTM — An open-source framework that provides extensible, unified implementations of mainstream knowledge tracing models (ranging from traditional psychometric approaches like Item Response Theory and BKT to deep neural implementations).
• Custom Deep Learning Repositories — If you prefer building custom neural network architectures (like DKVMN, SAKT, or GKT) using native frameworks, community collections like the Knowledge Tracing Collection on GitHub offer modular PyTorch building blocks trained on standard education datasets like ASSISTments.

To help narrow down the right approach, tell me:

• Are you looking for interpretable, probabilistic models (like Bayesian Knowledge Tracing) or high-capacity neural models (like Deep Knowledge Tracing)?
• What does your current data format look like (e.g., raw SQL logs, CSVs of correct/incorrect answers)?